Generated-source output writes ordinary C++ after llmcpp has replaced annotated requests with accepted implementations. It allows review without immediately building the final executable.
$ llmc++ --llm -fllm-offline main.cpp
This writes main.llm.cpp, using cached bodies and stopping before native compilation. An output filename ending in .cpp, .cc, or .cxx also selects generated-source mode, allowing an explicit destination with -o.
The rewritten file contains the implementation that conventional compilation will consume. Read it for edge cases, error handling, mutation, and dependency assumptions. It can also be compiled with an ordinary C++ compiler configured for the source's requirements.
Offline mode is useful during review because it requires the existing cached implementation rather than generating a replacement on a miss. Omit that setting only when fresh generation is intended.
Keeping the rewritten source produces an artifact independent of later generation exchanges. It does not establish behavioral correctness: the implementation still needs tests against the intended contract, just like manually written C++.